anomalyco/opencode
1
0
Fork 0
mirror of https://github.com/anomalyco/opencode.git synced 2026-07-22 02:05:26 -04:00
Code Issues 5.9k Packages Projects Releases 19 Wiki Activity

[PR #5051] feat: add experimental knowledge extraction compact method #11222

New Issue
Closed
opened 2026-02-16 18:16:01 -05:00 by yindo · 0 comments
yindo commented 2026-02-16 18:16:01 -05:00
Owner
Copy Link
Copy Source

Original Pull Request: https://github.com/anomalyco/opencode/pull/5051

State: closed
Merged: No

I feel that while working the agent discovers a lot of information writing beautiful overviews but promptly discards all of that when compacting. automatic /export to knowledge files and "compacting" by trimming the commentary would be ideal. This commit implements this approach, extracting knowledge (if needed) to ./.opencode/knowledge/ and then referencing it in the compacted message.

To enable the experimental feature:

  "compaction": {
    "extract_knowledge": true,
    "cleanup_transcripts": true,
    "auto_load_knowledge": false,
  }

Implementation by Claude Opus 4.5 (prototype).

Example of a created knowledge file:

---
created: 2025-12-04
source_sessions:
  - ses_51767658fffeSfJP0Ef3OZ3gyo
last_updated: 2025-12-04
---

# Knowledge Extraction Feature Architecture

## Overview

The knowledge extraction feature is triggered during session compaction (when context exceeds model limits). It extracts valuable, reusable information from sessions and persists it for future use.

## Configuration

Located in `packages/opencode/src/config/config.ts`, the compaction config schema includes:

| Option | Type | Default | Description |
|--------|------|---------|-------------|
| `extract_knowledge` | boolean | false | Extract and persist knowledge when compacting sessions |
| `cleanup_transcripts` | boolean | true | Remove transcript files after extraction |
| `auto_load_knowledge` | boolean | false | Automatically load all knowledge files into system prompt |

## Directory Structure

```
.opencode/
  sess/           # Session transcripts (exported before extraction)
  knowledge/      # Extracted knowledge files (*.md with frontmatter)
```

## Core Files

| File | Purpose |
|------|---------|
| `src/session/knowledge.ts` | Main orchestration - `SessionKnowledge.extract()`, `list()`, `load()`, `ensureDirectories()`, `parseKnowledgeReferences()` |
| `src/session/prompt/knowledge-extractor.txt` | System prompt defining what to extract and output format |
| `src/session/export.ts` | `SessionExport.toMarkdown()` and `writeToFile()` for transcript export |
| `src/session/compaction.ts` | Integration point - calls knowledge extraction during compaction |
| `src/session/system.ts` | `SystemPrompt.knowledge()` - loads knowledge files into context |
| `src/session/prompt.ts` | `extractKnowledgeReferences()` - parses `<knowledge_references>` blocks from summaries |
| `src/agent/agent.ts` | Built-in `knowledge-extractor` agent definition |

## Flow

1. Session transcript exported to `.opencode/sess/<session-id>.md`
2. `knowledge-extractor` agent analyzes transcript
3. Knowledge saved to `.opencode/knowledge/*.md` with frontmatter
4. Knowledge file paths included in compacted summary via `<knowledge_references>` blocks
5. On future prompts, referenced files loaded into system prompt

## Knowledge-Extractor Agent

Defined in `src/agent/agent.ts` with:
- **Tools**: read, write, edit, glob, grep (task, bash, webfetch disabled)
- **Permissions**: edit allowed, bash/webfetch denied
- **Prompt**: Loaded from `knowledge-extractor.txt`

## Knowledge File Format

```markdown
---
created: YYYY-MM-DD
source_sessions:
  - <session-id>
last_updated: YYYY-MM-DD
---

# Title

Content...
```

## What Gets Extracted

- Design decisions and architectural choices
- Technical specifications and schemas
- Bug resolutions and root causes
- Codebase learnings and patterns
- User preferences and coding style

## What is NOT Extracted

- Step-by-step debugging logs
- Raw tool outputs
- Verbose code dumps
- Transient discussion
- Routine operations
**Original Pull Request:** https://github.com/anomalyco/opencode/pull/5051 **State:** closed **Merged:** No --- I feel that while working the agent discovers a lot of information writing beautiful overviews but promptly discards all of that when compacting. automatic /export to knowledge files and "compacting" by trimming the commentary would be ideal. This commit implements this approach, extracting knowledge (if needed) to ./.opencode/knowledge/ and then referencing it in the compacted message. To enable the experimental feature: ```json "compaction": { "extract_knowledge": true, "cleanup_transcripts": true, "auto_load_knowledge": false, } ``` Implementation by Claude Opus 4.5 (prototype). Example of a created knowledge file: ````markdown --- created: 2025-12-04 source_sessions: - ses_51767658fffeSfJP0Ef3OZ3gyo last_updated: 2025-12-04 --- # Knowledge Extraction Feature Architecture ## Overview The knowledge extraction feature is triggered during session compaction (when context exceeds model limits). It extracts valuable, reusable information from sessions and persists it for future use. ## Configuration Located in `packages/opencode/src/config/config.ts`, the compaction config schema includes: | Option | Type | Default | Description | |--------|------|---------|-------------| | `extract_knowledge` | boolean | false | Extract and persist knowledge when compacting sessions | | `cleanup_transcripts` | boolean | true | Remove transcript files after extraction | | `auto_load_knowledge` | boolean | false | Automatically load all knowledge files into system prompt | ## Directory Structure ``` .opencode/ sess/ # Session transcripts (exported before extraction) knowledge/ # Extracted knowledge files (*.md with frontmatter) ``` ## Core Files | File | Purpose | |------|---------| | `src/session/knowledge.ts` | Main orchestration - `SessionKnowledge.extract()`, `list()`, `load()`, `ensureDirectories()`, `parseKnowledgeReferences()` | | `src/session/prompt/knowledge-extractor.txt` | System prompt defining what to extract and output format | | `src/session/export.ts` | `SessionExport.toMarkdown()` and `writeToFile()` for transcript export | | `src/session/compaction.ts` | Integration point - calls knowledge extraction during compaction | | `src/session/system.ts` | `SystemPrompt.knowledge()` - loads knowledge files into context | | `src/session/prompt.ts` | `extractKnowledgeReferences()` - parses `<knowledge_references>` blocks from summaries | | `src/agent/agent.ts` | Built-in `knowledge-extractor` agent definition | ## Flow 1. Session transcript exported to `.opencode/sess/<session-id>.md` 2. `knowledge-extractor` agent analyzes transcript 3. Knowledge saved to `.opencode/knowledge/*.md` with frontmatter 4. Knowledge file paths included in compacted summary via `<knowledge_references>` blocks 5. On future prompts, referenced files loaded into system prompt ## Knowledge-Extractor Agent Defined in `src/agent/agent.ts` with: - **Tools**: read, write, edit, glob, grep (task, bash, webfetch disabled) - **Permissions**: edit allowed, bash/webfetch denied - **Prompt**: Loaded from `knowledge-extractor.txt` ## Knowledge File Format ```markdown --- created: YYYY-MM-DD source_sessions: - <session-id> last_updated: YYYY-MM-DD --- # Title Content... ``` ## What Gets Extracted - Design decisions and architectural choices - Technical specifications and schemas - Bug resolutions and root causes - Codebase learnings and patterns - User preferences and coding style ## What is NOT Extracted - Step-by-step debugging logs - Raw tool outputs - Verbose code dumps - Transient discussion - Routine operations ````
yindo added the pull-request label 2026-02-16 18:16:01 -05:00
yindo closed this issue 2026-02-16 18:16:01 -05:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
acp
bad
bug
discussion
docs
documentation
enhancement
extension
good first issue
growth
help-wanted
mcp
model-problem
needs:compliance
nix
opentui
perf
pull-request
Mirrored from GitHub Pull Request
 tui
web
windows
zen
No Label pull-request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#11222
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?